How hogeheer499-commits/strix-halo-guide works
This repository is an independent Strix Halo local-LLM guide and evidence collection, not an inference server. Its main reader workflow is to choose a supported machine and configuration, prepare the operating system and GPU memory settings, install and qualify a runtime, then run a model and compare measured results. The guide also covers model/runtime route decisions, performance and stability, buyer guidance, Windows, and community evidence. Benchmark and analysis scripts support those evidence branches. The README documents the setup flow; the sampled scripts show specific benchmark behavior, but many guide subsystems are documentation or data whose implementation was not sampled.
Main entry points: README.md Quick start · data/README.md Evidence datasets · run-repro.py MTP repro · COMMUNITY_RESULTS.md Community results
Components 22
Setup Guide
- Quick start
README.md - Manual setup
STRIX_HALO_LOCAL_LLM_SETUP.md - Windows route
WINDOWS_START.md - Security guidance
SECURE_LOCAL_AI.md
Runtime Qualification
- Local LLM runtime
- Runtime qualification
RUNTIME_QUALIFICATION_2026-09-19.md - Model routes
CURRENT_MODELS.md
Benchmark Evidence
- Ollama comparison
data/raw/2026-07-16/ollama-0311-0312-0320-controlled/run-controlled-compare.py - ROCmFPX repro
data/raw/2026-07-16/step37-rocmfpx-q3-qualityplus/run-repro.py - MTP repro
data/raw/2026-08-09/qwen3-next-80b-mtp-b10330/run-repro.py - HIP control
data/raw/2026-08-09/qwen3-next-80b-mtp-b10330/run-hip-control.py - Thermal analysis
data/raw/2026-07-18/community-failsafe-corsair-thermal-sclk-issue24/analyze.py - Embedding check
data/raw/2026-07-25/nemotron-embed-1b-v2-official/run_embedding_check.py - Density analysis
scripts/parse_density_gate_campaign.py - Evidence charts
scripts/generate_charts.py - Evidence datasets
data/README.md - Benchmark method
BENCHMARKS.md
Community Evidence
- Community results
COMMUNITY_RESULTS.md - System evidence matrix
SYSTEM_EVIDENCE_MATRIX.md
Hardware Decisions
- Buyer guidance
BUYER_USE_CASES.md - Buyer validation
BUYER_PATH_VALIDATION.md
Other components
- Local AI user
Connections 18
- Local AI user → Quick start: follows
README.md - Quick start → Manual setup: offers manual route
README.md - Quick start → Windows route: offers Windows route
README.md - Quick start → Local LLM runtime: configures and runs
README.md - Quick start → Runtime qualification: checks qualification
README.md - Quick start → Security guidance: consults advisories
README.md - Local AI user → Buyer guidance: evaluates systems
README.md - Buyer guidance → Buyer validation: uses validation
- Local AI user → Benchmark method: reviews methods
README.md - Ollama comparison → Local LLM runtime: starts and queries
data/raw/2026-07-16/ollama-0311-0312-0320-controlled/run-controlled-compare.py - ROCmFPX repro → Local LLM runtime: starts and queries
data/raw/2026-07-16/step37-rocmfpx-q3-qualityplus/run-repro.py - MTP repro → Local LLM runtime: starts and queries
data/raw/2026-08-09/qwen3-next-80b-mtp-b10330/run-repro.py - HIP control → MTP repro: reuses repro
data/raw/2026-08-09/qwen3-next-80b-mtp-b10330/run-hip-control.py - Embedding check → Evidence datasets: prints metrics
- Thermal analysis → Community results: summarizes campaign
- Density analysis → Evidence datasets: parses campaign
- Evidence charts → Evidence datasets: visualizes results
- Community results → System evidence matrix: contributes system evidence